In [1]:

    

import keras as k
import keras.backend as b
import keras.layers as l
import numpy as np
import sys
import theano as t
import keras.initializations as i
sys.path.append('../src/mane/')


    

    




Using Theano backend.

In [2]:

    

import graph as g
fb = g.graph_from_pickle('../src/mane/data/egonets.graph')


    

In [3]:

    

def row_dot(inputs):
    return b.batch_dot(inputs[0], inputs[1], axes=[1,1])
def nce_loss(y_true, y_pred):
    y_true = y_true.reshape(y_pred.shape)
    return -b.log(b.sigmoid(y_pred * y_true))
def init_uniform(shape, name=None):
    return i.uniform(shape=shape, scale=1, name=name)


    

In [4]:

    

t_in = l.Input(batch_shape=(100,1), dtype='int32', name='target')
c_in = l.Input(batch_shape=(100,1), dtype='int32', name='class')


    

In [5]:

    

emb_t = l.Embedding(input_dim=len(fb), output_dim=200, 
                    name='target_emb', input_length=1,
                    init=init_uniform) (t_in)
emb_t = l.Reshape(target_shape=(200,))(emb_t)
emb_c = l.Embedding(input_dim=len(fb), output_dim=200, 
                    name='nce_emb', input_length=1,
                    init=init_uniform) (c_in)
emb_c = l.Reshape(target_shape=(200,))(emb_c)
emb_b = l.Embedding(input_dim=len(fb), output_dim=1,
                    name='nce_bias', input_length=1,
                    init='zero') (c_in)
emb_b = l.Reshape(target_shape=(1,))(emb_b)


    

In [6]:

    

dot_prod = l.Merge(mode=row_dot, output_shape=(1,), name='row_wise_dot')([emb_t, emb_c])
logits = l.Merge(mode='sum', output_shape=(1,), name='logits')([dot_prod, emb_b])
sigm = l.Activation('sigmoid', name='label')(logits)


    

In [7]:

    

model = k.models.Model(input=[t_in, c_in], output=sigm)


    

In [8]:

    

model.compile(loss='binary_crossentropy', optimizer='adam', name='mmodel')


    

In [9]:

    

model.summary()


    

    




____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to                     
====================================================================================================
class (InputLayer)               (100, 1)              0                                            
____________________________________________________________________________________________________
target (InputLayer)              (100, 1)              0                                            
____________________________________________________________________________________________________
nce_emb (Embedding)              (100, 1, 200)         807800      class[0][0]                      
____________________________________________________________________________________________________
target_emb (Embedding)           (100, 1, 200)         807800      target[0][0]                     
____________________________________________________________________________________________________
nce_bias (Embedding)             (100, 1, 1)           4039        class[0][0]                      
____________________________________________________________________________________________________
reshape_1 (Reshape)              (100, 200)            0           target_emb[0][0]                 
____________________________________________________________________________________________________
reshape_2 (Reshape)              (100, 200)            0           nce_emb[0][0]                    
____________________________________________________________________________________________________
reshape_3 (Reshape)              (100, 1)              0           nce_bias[0][0]                   
____________________________________________________________________________________________________
row_wise_dot (Merge)             (100, 1)              0           reshape_1[0][0]                  
                                                                   reshape_2[0][0]                  
____________________________________________________________________________________________________
logits (Merge)                   (100, 1)              0           row_wise_dot[0][0]               
                                                                   reshape_3[0][0]                  
____________________________________________________________________________________________________
label (Activation)               (100, 1)              0           logits[0][0]                     
====================================================================================================
Total params: 1619639
____________________________________________________________________________________________________

In [25]:

    

data_generator = fb.gen_walk('random_walk')


    

In [26]:

    

model.fit_generator(data_generator, nb_epoch=1, samples_per_epoch=3433150)


    

    




Epoch 1/1
   1768/3433150 [..............................]





    




---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-26-49126d574e58> in <module>()
----> 1 model.fit_generator(data_generator, nb_epoch=1, samples_per_epoch=3433150)

/usr/local/lib/python2.7/dist-packages/Keras-1.0.5-py2.7.egg/keras/engine/training.pyc in fit_generator(self, generator, samples_per_epoch, nb_epoch, verbose, callbacks, validation_data, nb_val_samples, class_weight, max_q_size, nb_worker, pickle_safe)
   1453                     batch_logs[l] = o
   1454 
-> 1455                 callbacks.on_batch_end(batch_index, batch_logs)
   1456 
   1457                 # construct epoch logs

/usr/local/lib/python2.7/dist-packages/Keras-1.0.5-py2.7.egg/keras/callbacks.pyc in on_batch_end(self, batch, logs)
     58         t_before_callbacks = time.time()
     59         for callback in self.callbacks:
---> 60             callback.on_batch_end(batch, logs)
     61         self._delta_ts_batch_end.append(time.time() - t_before_callbacks)
     62         delta_t_median = np.median(self._delta_ts_batch_end)

/usr/local/lib/python2.7/dist-packages/Keras-1.0.5-py2.7.egg/keras/callbacks.pyc in on_batch_end(self, batch, logs)
    186         # will be handled by on_epoch_end
    187         if self.verbose and self.seen < self.params['nb_sample']:
--> 188             self.progbar.update(self.seen, self.log_values)
    189 
    190     def on_epoch_end(self, epoch, logs={}):

/usr/local/lib/python2.7/dist-packages/Keras-1.0.5-py2.7.egg/keras/utils/generic_utils.pyc in update(self, current, values, force)
    117                 info += ((prev_total_width - self.total_width) * " ")
    118 
--> 119             sys.stdout.write(info)
    120             sys.stdout.flush()
    121 

/usr/local/lib/python2.7/dist-packages/ipykernel/iostream.pyc in write(self, string)
    315 
    316             is_child = (not self._is_master_process())
--> 317             self._buffer.write(string)
    318             if is_child:
    319                 # newlines imply flush in subprocesses

ValueError: I/O operation on closed file

In [20]:

    

next(data_generator)


    

    
Out[20]:





({'class': array([1487, 1408, 2222,  136, 1927, 2540, 1792, 1004, 2011, 2624, 1447,
         2510, 1195, 3385, 3184, 2735, 1148], dtype=int32),
  'target': array([526, 526, 526, 526, 526, 526, 526, 526, 526, 526, 526, 526, 526,
         526, 526, 526, 526], dtype=int32)},
 {'label': array([ 1.,  1.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,
          0.,  0.,  0.,  0.], dtype=float32)})

In [21]:

    

la = next(data_generator)


    

In [24]:

    

len(la[0]['class'])


    

    
Out[24]:





17

In [ ]: